I’ve also stumped upon the missing size when I tried to use it in crossover/2 as p1.size instead of recalculating length(p1.genes) at each iteration. I took a while to figure out it was not retained after initializing the chromosome’s population.
About the chromosome update you mention, we can discuss if you really want the age to increase after a crossover (when you get a kind of new chromosome) or if we should consider a new generation only after a mutation. But I think this are not so relevant aspects in respect of the main code’s objective.
Similar consideration goes for the silly optimization I made in evaluate/3, when I opted to use the precalculated fitness for the last sorting step:
...
|> Enum.sort_by(& &1.fitness, &>=/2)